Experimentally Validated hERG Pharmacophore Models as

Subscriber access provided by Northern Illinois University

Article

Experimentally validated hERG pharmacophore models as cardiotoxicity prediction tools Jadel M. Kratz, Daniela Schuster, Michael Edtbauer, Priyanka Saxena, Christina E. Mair, Julia Kirchebner, Barbara Matuszczak, Igor Baburin, Steffen Hering, and Judith M. Rollinger J. Chem. Inf. Model., Just Accepted Manuscript • DOI: 10.1021/ci5001955 • Publication Date (Web): 22 Aug 2014 Downloaded from http://pubs.acs.org on September 3, 2014

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Journal of Chemical Information and Modeling is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Journal of Chemical Information and Modeling

Experimentally validated hERG pharmacophore models as cardiotoxicity prediction tools Jadel M. Kratz,†,§ Daniela Schuster,*,‖ Michael Edtbauer,‖ Priyanka Saxena,⊥ Christina E. Mair,§ Julia Kirchebner,‖ Barbara Matuszczak,‖ Igor Baburin,⊥ Steffen Hering,⊥ Judith M. Rollinger*,§ †

Departamento de Ciências Farmacêuticas, Universidade Federal de Santa Catarina, 88.040-900,

Florianópolis, SC, Brazil. §

Institute of Pharmacy/Pharmacognosy and Center for Molecular Biosciences Innsbruck,

University of Innsbruck, Innrain 80-82, 6020 Innsbruck, Austria. ‖

Institute of Pharmacy/Pharmaceutical Chemistry and Center for Molecular Biosciences

Innsbruck, University of Innsbruck, Innrain 80-82, 6020 Innsbruck, Austria. ⊥ Department

of Pharmacology and Toxicology, University of Vienna, Althanstraße 14, 1090

Vienna; Austria. * corresponding authors

ACS Paragon Plus Environment

1


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 2 of 51

ABSTRACT:

The goal of this study was to design, experimentally validate, and apply a virtual screening workflow to identify novel hERG channel blockers. The hERG channel is an important antitarget in drug development, since cardiotoxic risks remain as a major cause of attrition. A ligand-based pharmacophore model collection was developed and theoretically validated. The seven most complementary and suitable models were used for virtual screening of in-house and commercially available compound libraries. From the hitlists, fifty compounds were selected for experimental validation through bioactivity assessment using patch clamp techniques. Twenty compounds inhibited hERG channels expressed in HEK 293 cells with IC50 values ranging from 0.13 to 2.77 µM, attesting the suitability of the models as cardiotoxicity prediction tools in a preclinical stage.


2

Page 3 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


INTRODUCTION The human ether-a-go-go-related gene (hERG) channel plays a critical role in cardiac action potential repolarization. Its importance was amplified with the association of drug-induced hERG blocking with an increased incidence of a fatal type of arrhythmia, named “torsades de pointes” (TdP).1-3 In fact, many drugs, including non-cardiac drugs, such as antimicrobials, neuroleptics, antipsychotics, antiarrhythmics and antihistamines, have been withdrawn from the market or received severe restrictions to use due to hERG-related cardiotoxicity (Chart 1).1,2 With repercussions in both drug discovery and clinical practice, the hERG channel has been recognized as a primary antitarget in the screening of drug candidates.4,5 A number of preclinical models have been developed to assess potential pro-arrhythmic properties.6 hERG liability assessment and management became an important part of every project in the pharmaceutical industry. However, despite all the efforts and progress obtained in the evaluation of compounds effects on hERG channels, QT prolongation remains a major cause of attrition during current drug development.7 Chart 1. Representative structures of drugs withdrawn from the market due to hERG-related QT interval prolongation and severe risk of fatal arrhythmias.


3


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 4 of 51

Several issues should be recognized and investigated in the preclinical setting in order to fully assess the hERG liability of a new chemical entity. The complexity of the ventricular repolarization,3 along with the channel promiscuity, potential interactions with other ion channels, different binding modes, and modes of action,8-10 are just a few issues that add up to the multifaceted cardiotoxicity prediction arena. Due to high costs and limitations of available preclinical models,11 an integrated risk assessment seems desirable. Association of in silico and experimental approaches has been suggested as the cornerstone of a cost-effective and successful preclinical evaluation of the cardiotoxic risks of drug candidates.6,7 Many molecular modeling studies focused on hERG block prediction have been published, covering QSAR12-15 and rule-based models,16 pharmacophores,17-21 structure-based studies employing homology models9,22, machine learning,23,24 and computational approaches describing the physiology of the electrical wave propagation in the heart.25 Recent advances in the prediction of hERG blockage have been summarized by Wang et al.26 Nonetheless, while in silico models have great potential, particularly in early drug development, only few examples of experimentally validated models comprising publicly available data are available in the literature.15,17,27-31 With the increasing need for reliable cardiotoxicity prediction tools, especially on hERG blocking and its QT prolongation potential, we designed and experimentally validated a set of complementary ligand-based pharmacophore models. This study focused on pharmacophore models, because in the safety profiling field many pharmacophore models have been established and generally provide satisfactory results.32,33 The predictive tools developed and validated in this study were able to accurately and prospectively predict hERG blocking by novel compounds.


4

Page 5 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


RESULTS AND DISCUSSION The general workflow of this study is depicted in Figure 1. hERG blockers databases were assembled from literature and used for ligand-based hERG blocker pharmacophore models generation with the two software packages Discovery Studio – Catalyst and LigandScout. The models were theoretically validated using literature datasets and a decoy set. Only models with high enrichment factors and complementary virtual hits (VH) were employed for prospectively screening compound libraries. Biological evaluation of each model was carried out by selecting 513 VH for experimental validation using a patch clamp assay. All models - except for one successfully identified previously unknown hERG blockers from diverse chemical classes.


5


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 6 of 51

Figure 1. General workflow employed in this study.


6

Page 7 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Pharmacophore modeling. As reported before, pharmacophore generation algorithms and virtual screening (VS) protocols yield different pharmacophore models and hit lists, even if the training molecule(s) for model generation are identical.34,35 Remarkably, all pharmacophore programs are able to prospectively find assorted active hits, and pharmacophore algorithms can be combined together in order to increase the success of hit compound identification.34,35 Our goal was the achievement of a consensus set of models26,36 and its application in a prospective campaign, and not the direct comparison of different softwares, model generation approaches, or training datasets. Therefore, two pharmacophore modeling software packages were used for modeling and VS. As a further step, aiming at high confidence ligand-based models, only compounds with results from functional assays, e.g. patch clamp, were included during database compilation, reducing considerably the size of the training sets. Even though, on average, the correlation between hERG functional and binding assays is moderate in ChEMBL,30 the preparation of datasets with only high quality data was considered a key step toward the development of reliable models.26 On the one hand, a Catalyst pharmacophore model37 was generated mostly based on clinically used drugs. This model was designed to recognize a broad variety of hERG blockers from the literature. On the other hand, a parallel pharmacophore modeling approach38 was pursued using LigandScout.39 In this workflow, several pharmacophore models recognizing a subset of active compounds from the ChEMBL database, and being restrictive against inactive and decoy molecules, were combined. The screening model set was assembled so that the majority of active compounds was correctly found, while the overall number of false positive predictions remained small. The Catalyst pharmacophore model was based on 18 chemically diverse hERG blockers from the literature, all evaluated with the same patch clamp assay (Chart 2).15 The generated model was


7


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 8 of 51

subsequently validated using a test set of 19 compounds compiled from previously published studies (Chart 3).15,40-49

Chart 2. Training compounds used for Catalyst model generation.a

a

Compounds 1, 3 and 4 were also part of the training set.


8

Page 9 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Chart 3. Test compounds used for Catalyst model theoretical validation.


9


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 10 of 51

Among the ten models generated by the program, the one recognizing the most active hits from both the training and test sets was selected for further studies (Figure 2).

Figure 2. Catalyst pharmacophore model for hERG blockers. The distances between the features are indicated in Å. Chemical features are color-coded: positively ionizable – red, hydrophobic – cyan, aromatic hydrophobic – blue.

In this theoretical validation, the Catalyst model was able to find 14 out of 18 compounds in the training set and 8 out of 19 test set compounds (Table 1 and Supporting Information Table S1). Table 1. Test set screening results using the Catalyst model. Cpd Experimental IC50 [µM] Found by Model Reference X

49

21

0.14

22

0.17

23

0.2

24

0.81

43

25

1.1

41

40

X

46


10

Page 11 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


X

15

26

1.3

27

1.39

28

1.5

29

1.9

30

3.3

31

3.4

32

3.73

X

43

33

4.6

X

44

34

5.95

43

35

7.2

48

36

7.8

47

37

10

15

38

12.1

43

39

17.3

15

X

42 15

X

15 45

X

49

The pharmacophore model collection generated with LigandScout comprised six models, which were validated using a set of 86 highly active hERG blockers collected from literature and a decoy set (12771 drug-like decoys supposed to be hERG inactive). Each model was based on two training compounds (Figure 3). The refinement of each automatically generated model is described in detail in the Supporting Information.


11


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 12 of 51

Figure 3. Training compounds and optimized ligand-based LigandScout pharmacophore models (M1 – M6). Chemical features of the models are color-coded: hydrophobic – yellow; hydrogen bond acceptor – red; aromatic ring – blue parallel rings; positively ionizable group – blue star; exclusion volumes – grey. In general, most models were composed by the classical pharmacophore features for hERG blockers reported by other groups: two to three hydrophobic features, some of them grouped with an aromatic ring.15,21,50 All models but M2 incorporated one or two hydrogen bond acceptors. Finally, four models (M3 – M6) included a positively ionizable group. In comparison, the Catalyst


12

Page 13 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


model also fulfilled these general features with three hydrophobic / aromatic features and a positively ionizable group. Due to the unique shape of the binding site and hydrophobic character, the hERG channel has been shown to interact with a wide range of chemical structures. But, in general, it is recognized that one or two hydrophobic groups interact with Phe656 side chains, and a basic nitrogen (protonated at biological conditions), or aromatic ring, is involved in cation–π interaction with Tyr652 residues.50 In this view, the features present in our models largely represent the classical chemical moieties present in known hERG blockers. The performance of each model alone and also as a single group (parallel screening) is shown in Table 2. The receiver operating characteristic (ROC) plot of the database screenings with the combined model set is given in Figure 4.

Table 2. Performance of LigandScout models in single and parallel screening against the hERG highly active database and the decoy set. Model

True positive hitsa False positive hitsb EFc

ROC-AUCd

M1

19 (22.09%)

20 (0.15%)

72.8

0.61

M2

13 (15.11%)

2 (0.016%)

129.6 0.58

M3

26 (30.23%)

58 (0.45%)

46.3

0.65

M4

20 (23.25%)

140 (1.10%)

18.7

0.61

M5

7 (8.13%)

2 (0.016%)

116.3 0.54

M6

24 (27.90%)

202 (1.58%)

15.9

0.63

M1 – M6e 73 (84.88%)

400 (3.13%)

23.1

0.91

a

Virtual hits from the highly active database (total of 86 compounds). bVirtual hits from the decoy set (total of 12771 compounds). cThe enrichment factor (EF) measures the yield of actives proportionally to the ratio of actives in the database. The maximum theoretical EF for the data sets


13


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 14 of 51

used in this study is 148.5. dReceiver Operating Characteristic (ROC) area under the curve (ROCAUC); Values above 0.5 indicate yields better than random selection of hits (see experimental section for a detailed description of models performance assessment). eVirtual screening performed with all models simultaneously – parallel screening.

Figure 4. ROC plots of screening the highly active database and decoy set using the models M1 – M6 in parallel (ROC-AUC = 0.91) (A) and using the Catalyst model (ROC-AUC = 0.89) (B). A truly valid group of pharmacophore models should be able to differentiate between active and inactive molecules, and perform complementarily. The theoretical validation results showed that this gold standard was achieved for M1 – M6. In the parallel screening, the models were able to cover 84.9% of truly active compounds (ranging from 8 to 30% as single models) with few false positive hits giving a ROC-AUC value of 0.91. These results indicated a remarkable performance of the models in combination. In a single model appraisal, M3 showed the highest true positive hit rate, while M2 was the most specific model. On the other hand, M4 and M3 were the most


14

Page 15 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


complementary models, while M5 predicted only 4 unique hits. For comparison, the Catalyst model was screened against the data sets used for models M1-M6. The model retrieved 1287 hits in the decoy set and 68 out of 86 highly active hERG blockers. Although its performance in finding active compounds was nearly comparable to the combined models M1-M6, it had a much higher rate of false positive hits as reflected by the EF of 7.5 (vs. 23.1 for M1-M6). To additionally validate the generated models and the VS strategy by an independently assembled data set, all active compounds discussed in an excellent review by de Bruin and coworkers51 were submitted to an activity classification (Table 3). Table 3. Retrospective validation results using the external dataset as reported by deBruin et al.51

Cpd

IC50 [µM]a

LigandScout models

Catalyst predicted active

LigandScout predicted active M1 M2 M3 M4 M5 M6

astemizole (5)

0.0009

X

X

X

cisapride (3)

0.002

Xb

X

terodiline

0.004

dofetilide

0.005

ibutilide

0.01

X

X

sertindole (4)

0.014

X

X

X

pimozide (9)

0.015

X

X

X

terfenadine (1)

0.02

X

haloperidol (7)

0.027

X

X

thioridazine (8)

0.033

X

X

almokalant

0.05

X

X

X

X X X

X X


15


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

azimilide

0.1

X

verapamil (21)

0.14

X

risperidone (11)

0.15

X

domperidone

0.16

X

loratadine (22)

0.173

aprindine

0.23

sparfloxacin

0.23

olanzapine (12)

0.231

X

ebastine

0.3

X

quinidine (33)

0.3

X

mibefradil

0.35

X

X

X

mizolastine

0.35

X

X

X

propafenone

0.44

X

bepridil

0.6

X

amiodarone

1.0

X

tamoxifen

1.0

X

desipramine (27)

1.39

chlorpheniramine

1.6

X

disopyramide

1.8

X

ketoconazole (29)

1.9

tedisamil

2.5

fluoxetine

3.1

imipramine (31)

3.4

flecainide

3.91

amitriptyline (37)

4.66

fexofenadine

5.0

Page 16 of 51

X

X

X

X

X

X

X

X

X

X

X

X

X

X

X


16

Page 17 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


mefloquine

5.6

nitrendipine

10

diltiazem (39)

10

cibenzoline

23

sematilide

25

grepafloxacin

27

diphenhydramine

30

clarithromycin

32.9

erythromycin

72.2

sotalol

100c

cetirizine

108

nifedipine

275

procainamide

310

ciprofloxacin

966

X

X

X

X

X

X

74

phenytoin

X

X

X

X

a

IC50 values as reported in the original publications given in the review of deBruin et al.51 bSince this dataset was assembled completely independently from this study, some compounds match the training set used for model development, and are marked in grey. cCompounds with IC50 values > 100 µM were considered inactive. Overall, 65% of the compounds were found by the combination of the two different software models (34 out of 52 compounds). The Catalyst model alone was able to find 58% of the compounds, while the LigandScout models showed an inferior performance, finding only 37% of the external dataset of hERG blockers. Both programs found one inactive compound, cetirizine. The hit lists presented some overlap, with 15 consensus hits, but Catalyst and LigandScout models found 15 and 4 unique hits, respectively. Restricting the analysis to potent hERG blockers with IC50 values ≤ 1 µM, the combination of models was able to find 89% of the highly active blockers (24 out of 27 compounds). The performance of both models improved in this subset: the Catalyst


17


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 18 of 51

model prediction rate increased from 58% to 85% (23 out of 27 compounds), and the LigandScout models were able to find 44% of the highly active compounds, with one additional unique hit. In this view, the global results obtained in the theoretical validation indicated that the main modeling objective was successfully accomplished. Complementary models were constructed, retrieving a good coverage of active compounds (average 63.50% true hit rate) combined with satisfactory selectivity (average 0.55% false positive hit rate). Therefore, we decided to use the models from both softwares for the prospective virtual screening of compounds libraries in the search of putative hERG blockers to be subjected to experimental testing and for validation of the generated models. Prospective Virtual Screening of In-House and Commercial Databases. The pharmacophore models were experimentally validated using compounds available from the in-house library of the Institute of Pharmacy at the University of Innsbruck or provided by the commercial supplier SPECS (www.specs.net). The primary goal was to obtain, for each of the pharmacophore models, a sufficient number of compounds predicted to be active. Initially, a prospective VS was carried out against the in-house 3D molecular database (In-house, 3986 compounds), which led to 174 and 93 VH for LigandScout models (M1 – M6) and Catalyst models, respectively (average hit rate of 3.4%). Because the hitlists presented a restricted number of chemical scaffolds, and did not cover all the models properly, the SPECS synthetic products (SPECS SP, 202,907 compounds) and SPECS natural products (SPECS NP, 453 compounds) databases were additionally screened. This endeavor retrieved more than 20,000 VH (average hit rate 6.5%). The combined hitlists were clustered by structural diversity and inspected for the removal of compounds with known hERG activity. In total, 50 compounds were selected for biological evaluation with 5 to 13 compounds per model. As selection criteria we used the chemical diversity, high geometric fit-score, consensus


18

Page 19 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


hits, and confirmed purity and identity (all tested compounds showed a purity ≥ 95%, assessed by HPLC-MS). A chart with all the structures is available in the Supporting Information (Chart S1). Experimental Validation Results. In a first screening, employing a two micro electrode voltage clamp assay on hERG channels expressed in Xenopus oocytes, all compounds were applied at a concentration of 30 μM.52 Thioridazine (8), a known hERG blocker,15 was employed as positive control. A 30% reduction of peak tail current was established as the cut-off for positive hERG blockade. This campaign identified twenty hERG blockers that showed inhibitions between 32.3 and 78.9% (Figure 5). Complete screening data including models coverage and compounds sources is available in the Supporting Information (Table S2).

Figure 5. Inhibition by compounds (30 μM) of hERG current in the oocyte two micro electrode assay, given as mean ± SE (n = 3-5). Thioridazine (8) was used as positive control.


19


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 20 of 51

For further in vitro characterization, a more restraining cut-off was employed. Only the 13 compounds that showed ≥ 50% block in the preliminary screening were selected for IC50 determination in a patch clamp assay in HEK293 cells.53,54 Figure 6 shows the concentrationinhibition curves and the structures of the respective compounds. This study revealed a potent concentration-dependent block of hERG currents by all the compounds, with IC50 values ranging from 0.13 to 2.77 μM (Table 4). Table 4. Inhibitory activity of selected compounds in the HEK 293 cells expressing hERG channels in the planar patch-clamp assay. Cpd IC50 [µM]a

a

Predictionb

51

0.13 ± 0.02 M3

52

0.15 ± 0.03 M1

53

0.30 ± 0.08 M6

54

0.42 ± 0.07 M6

55

0.58 ± 0.24 M5

56

0.64 ± 0.14 M4, Catalyst

57

0.67 ± 0.10 Catalyst

58

0.70 ± 0.28 M4, Catalyst

59

0.96 ± 0.09 Catalyst

60

0.99 ± 0.22 M1, M6

61

1.61 ± 0.14 M4, Catalyst

62

1.92 ± 0.69 M1, Catalyst

63

2.77 ± 0.61 M6

IC50 ± SE (n = 3-5). bVirtual hit of the depicted model(s) in the prospective virtual screening.


20

Page 21 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Figure 6. Concentration-response curves of hERG current inhibition by compounds in the HEK293 cells patch clamp assay (n = 3-5). Only compounds that showed ≥ 50% block in the preliminary screening were selected for this further validation step.


21


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 22 of 51

Comparison of prospective model performance. From the seven ligand-based pharmacophore models developed and investigated in this study, six successfully passed the theoretical and experimental validation. A final true hit rate of 40% (20 out of 50 compounds selected for biological screening) was obtained, including ten compounds active in the submicromolar range (Table 4, Figure 6). Regarding complementarity, 33% of the consensus hits and 50% of the single hits were active. Table 5 summarizes the prospective prediction performance. A structure similarity search within the highly active database (Tanimoto coefficient 0.5), using the new true hits as queries, revealed that only 5 out of 13 molecules retrieved results, corroborating the scaffold hopping potential of the models (data not shown). On the other hand, although structurally diverse from the classical hERG blockers, the newly identified hERG blockers covered a part of the physicochemical drug space already populated by known hERG blockers (chemical space analysis available in the Supporting Information, Figure S1). Table 5. Individual models performance evaluation on the basis of the experimental results. Model

Tested cpdsa

Active cpdsb

Success rate

M1

8

3

37.5%

M2

5

0

00.0%

M3

6

2

33.3%

M4

11

5

45.5%

M5

6

2

33.3%

M6

9

6

66.7%

Catalyst

13

8

61.5%

a

Number of predicted compounds tested, considering consensus hits. bNumber of compounds that reached at least 30% reduction of hERG peak tail current in the patch clamp assay in Xenopus oocytes.


22

Page 23 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


LigandScout models M4 and M6 as well as the Catalyst model showed superior performance, thus, they can be considered as particularly reliable tools. The negative highlight is M2; despite its good theoretical validation, all five virtual hits tested in this study were inactive. This pharmacophore was the only model that did not include any hydrogen bond acceptor feature, and this discrimination might be linked with the poor experimental performance. Some studies have underlined the importance of hydrogen bond acceptor features for appropriate fitting of molecules into the hERG channel, especially for uncharged hERG blockers.18,36,50 The prospectively achieved rate of true positive hits (success rate, Table 5) did not correlate with the enrichment rates observed in the retrospective validation (Table 2). M2, the model with the highest EF, did not find any new hERG inhibitors in the prospective screening. However, M6 and the Catalyst model, both with the lowest EFs, showed superior success rates in the experimental validation. Generally, enrichment metrics highly depend on the data set(s) used for model validation. Therefore, the metrics calculated for only one data set may not adequately represent the actual predictive power of a model. Thus, for a broader picture on the models’ performances, additional, independently assembled, publicly availably hERG data sets provided by Wang et al. (http://cadd.suda.edu.cn/admet/downloads/hERG)24 were screened (Table 6).

Table 6. Screening of the LigandScout models M1-M6 and the Catalyst model against four external data sets provided by Wang et al.24


23


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 24 of 51

Data set, M1 M2 M3 M4 M5 M6 M1Cataly composit (%EFm (%EFm (%EFm (%EFm (%EFm (%EFm M6 st (%EFm (%EFm ion ax) ax) ax) ax) ax) ax) ax) ax) (actives / decoys)

Comm on hits of M1M6 vs. Cataly st

training Wang

9/4 (69.2)

6/0 (100)

34/5 (87.2)

17/2 (89.5)

5/1 (83.3)

8/4 (66.7)

71/15 (82.6)

141/36 (79.7)

42

test Wang (61/59)

6/0 (100)

1/0 (100)

5/0 (100)

2/0 (100)

0/0 (-)

2/0 (100)

13/0 (100)

33/8 (80.5)

10

WOMB AT Wang

0/0 (-)

0/0 (-)

7/1 (87.5)

3/1 (40.9)

2/0 (100)

2/0 (100)

12/2 (85.7)

31/0 (100)

8

2/2 (50)

1/7 (12.5)

16/13 (55.2)

0/2 (0)

0/13 (0)

19/33 (36.5)

61/98 (38.4)

15

(303/300 )

(55/11) Pubchem 0/1 (0) Wang (250/169 3)

Similar to the previous validation screenings, all models performed well in terms of enriching active compounds in their hit lists. One data set, the PubChem data set, was the most challenging one of the screened databases. Some of the LigandScout models were not able to find a single active compound in this data set. However, the PubChem data were generated using a fluorescence-based hERG block assay, in which assay interferences by chromophoric compounds cannot be excluded. Additionally, this assay clearly differs from the “gold standard” patch-clamp used as starting point for our modeling data sets and experimental validation of virtual hits. Also Wang et al. retrieved their lowest general accuracy metrics with the PubChem data set.


24

Page 25 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Importantly, the LigandScout models M1-M6 and the Catalyst model did not find exclusively identical hits (Table 6). On average, 30% of the LigandScout model hits and 72% of the Catalyst model hits showed no overlap. The reasons for the diverging hits lie in the different chemical feature definitions and screening algorithms, as discussed by Spitzer et al.35 and Sanders et al.34 In all of the screening databases (Table 6), the poor prospective performance of M2 and the superior performances of M6 and the Catalyst model were not reflected. Anyway, it has to be considered that only up to 13 compounds were tested per model. For M2, only 5 hits were evaluated in vitro because of its high restrictivity. Generally, virtual databases for model validation are carefully assembled and seeded with structurally diverse, known active compounds. However, in a prospective screening, there is no pre-information available if or how many active molecules from which chemical classes can be identified. It can therefore happen that a retrospectively excellently validated model doesn’t retrieve new active hits from a prospectively used screening database. The reported parallel screening approach aimed at a preferably complete retrieval of hERG blockers (high sensitivity) and a correct classification of inactive compounds (high specificity). The latter was achieved with very low false positive hit rates in all the investigated databases. The specificity was even superior to the machine learning models reported by Wang et al. The sensitivity however could still be improved to identify also smaller and chemically more diverse true positive hits. For example, one of the (aromatic) hydrophobic features from the Catalyst model could be set optional. This small change in the model would immediately considerably improve the true positive hit rate, e.g. enabling highly active compounds like terodiline to map all required features. However, at the same time, the specificity of the model and, in turn, its EFs would


25


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 26 of 51

decrease. This would lead to an unacceptable amount of virtual hits (average 30% of database entries), hindering the focus on most probable hERG blockers. Regarding the relevance of the novel hERG blockers found in this study, among the hits are neuroleptic drugs (azaperone, 51), intermediary products in organic chemistry (phenodianisyl, 52), local anesthetics (cinchocaine, 54), and natural products. Even though the association between cardiotoxicity and in vitro hERG blocking is sometimes misleading, some authors already suggest “cardiac safety margins” of at least 30 fold between the ED50 of a drug and its IC50 on the hERG channel. An extensive evaluation of 100 drugs done in 2003 by Redfern et al. supports the safety margin of a 30 fold for the treatment of serious illnesses. Further they suggest a safety margin of >100 fold for the treatment of less serious illnesses while a safety margin of >10 fold may be acceptable for a medication of life threatening diseases. Voacangine (53), which showed an IC50 value of 0.30 ± 0.08, is an iboga alkaloid found in the rootbark of Voacanga africana and other species.55 This compound serves as a precursor for the semi-synthesis of ibogaine, a not licensed anti-addictive iboga alkaloid. The potent hERG blocking effect of ibogaine has been published recently,55,56 after we selected its close analog, voacangine, from our virtual hit list as hERG blocking candidate for biological testing. Although the predictive power of most reported models was high, the use of these models for predictive screening studies needs to be performed with caution. The theoretical model validation already showed that the models could not correctly identify 100% of the active compounds from the literature. As in many virtual screening approaches, model optimization here is a balance act between sensitivity and specificity.57 For the hERGscreen project, it was more important to find


26

Page 27 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


active compounds with comparably low experimental efforts. Therefore, the more restrictive models were favored. However, when screening for potentially toxic effects, it is desirable to detect all relevant compounds. The models can therefore mainly be used to prioritize compounds likely to be active hERG blockers, since a compound not fitting into the models may still be active. The reported model collection presented in this study will be even more useful when complemented by other virtual classification tools. For instance, models developed based on machine learning methods, which employ inactive compounds in large numbers, are likely to cover a broader chemistry property space and represent good candidates for balancing our approach.23,24,26,58

CONCLUSION Several pharmacophore models have been developed in this study using two different software tools. Based on a theoretical validation, seven models have been qualified with high predictive power. Experimental validation further attested the Catalyst model and five of the six generated LigandScout models for application in a preclinical stage to identify putative hERG channel blockers from large data sets. The experimental confirmation that the models developed in this study can accurately predict hERG block by previously unknown compounds of natural origin is of special interest. These models (the Catalyst model together with a parallel screening in LigandScout with all models excluding M2) are currently implemented as prediction tools in an ongoing EU-project (hERGscreen, 295174, www.uibk.ac.at/pharmazie/pharmakognosie/hergscreen), where we aim at


27


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 28 of 51

a target-oriented identification and isolation of hERG channel blockers from commonly consumed botanicals. The repercussion of hERG blockage for drug candidates and its awareness is reflected in the guideline on the non-clinical strategy to unmask the QT prolonging potential by human pharmaceuticals published by the ICH (International Conference on Harmonization) in 2005 (ICH S7B Guideline).5 This regulation, however, has never been systematically applied on botanicals. As natural products, such as dietary supplements, spices, and herbal medicinal products, continue to increase in popularity, there is an urgent need for studies aimed to critically assess their potential cardiotoxic risks. Highly potent hERG blockers present in these products, even in small amounts, can threaten the “cardiac safety margins” proposed by some authors.4 The EU-project’s efforts accordingly aim at determining, identifying, understanding and ultimately reducing the cardiotoxic safety liabilities of frequently used botanicals. The pharmacophore models generated and validated in this study now offer the possibility to mine large natural product databases to prioritize botanicals with putative hERG blocking activities to be phytochemically investigated in detail and subjected to focused biological testing.

EXPERIMENTAL SECTION Hardware and Software. In silico studies were carried out on work stations running Windows Vista, Windows 7, and/or Linux CentOS 5. Pharmacophore modeling and VS experiments were performed using LigandScout 3.03b (Inte:Ligand, Vienna, Austria),39 Catalyst 4.7 and Discovery


28

Page 29 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Studio 3.0 (Accelrys Inc.: San Diego, CA, USA, 2001-2013). Structural conformational analyses were performed in OMEGA version 2.3.3,59-61 as incorporated in LigandScout, and Catalyst 4.7 or Discovery Studio 3.0. Structural clustering was calculated using Discovery Studio 3.0. Structural drawing was performed in ChemBioDraw Ultra 12 (CambridgeSoft Corp., Cambridge, MA, USA). Databases Compilation and Preparation. LigandScout Models. For pharmacophore model development in LigandScout, databases of published compounds with known hERG-blocking activities were assembled. As data source, the ChEMBL platform62 (version 12) was used. All compounds with annotated hERG activity data were analyzed. Only hERG blocking activities determined using patch clamp techniques63 were kept. From the recent literature,64-68 six additional compounds were added to the data set resulting in a database of 609 entries. Each molecular structure and activity annotation was checked for correctness in the original literature. This data set was split into three sections: (i) highly active hERG blockers with IC50 ≤ 1 µM (n = 141), (ii) active compounds with IC50 values between 1 and 100 µM (n = 448), and (iii) inactive molecules with IC50s > 100 µM (n = 20). In order to remove redundant information on structurally similar compounds from the same activity class, each set was clustered by molecular diversity using FCFP6 fingerprints available in Discovery Studio. Further on, by visual inspection of the clusters, only the most diverse molecules were kept. The final highly active set comprised 86 compounds, the active set 206 structures, and the inactive set 19 molecules. These data sets were converted into 3D multiconformational data bases using OMEGA as incorporated in LigandScout. The BEST settings were used, calculating up to 400 conformers per molecule. Catalyst Data Set. For model generation and validation, a literature data set comprising 37 drugs was assembled, with a focus on clinically used hERG blockers. The compounds were generated


29


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 30 of 51

with the molecule builder in Catalyst 4.7 and submitted to conformational analysis using the BEST settings with a maximum number of 250 conformers per molecule and an energy range of maximal 20 kcal / mol above the calculated energy minimum. Decoy Set. For theoretical validation of LigandScout models, a previously reported decoy set was used.69 Decoys are molecules that most likely do not show any binding affinity to the target, but their bioactivity has not yet been tested. 3D conformations of the structures were calculated using OMEGA, incorporated into LigandScout, using FAST configuration. 100 conformations were calculated for each structure. The result was an optimized 3D multiconformational screening database with 12771 molecules. External Data Set. For the retrospective theoretical validation of all models, an external database comprising 52 known hERG blockers was compiled. This dataset was previously published by de Bruin and coworkers.51 For all compounds a maximum of 100 conformers were computed with BEST settings using OMEGA, incorporated into LigandScout, or Discovery Studio. SPECS and In-house Databases. Both synthetic (SPECS SP) and natural products (SPECS NP) collections were downloaded from the company’s website (Nov/2012 – www.specs.net). The structures from compounds available in-house were drawn at ChemBioDraw. 3D multiconformational screening databases were created using OMEGA, incorporated into LigandScout, or Discovery Studio. For each molecule entry, a maximum of 100 conformers was computed with BEST settings, except for SPECS NP, in which 400 conformers were computed. The final SPECS SP, SPECS NP and In-house databases consisted of 202,907, 453 and 3,986 unique entries, respectively.


30

Page 31 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Ligand-based 3D Pharmacophore Models Generation. LigandScout Models. The highly active ChEMBL database (86 compounds) was loaded into the espresso module of LigandScout set with default configuration and exclusion volumes coating. This module employs the molecular alignment algorithm for shared feature pharmacophore model generation.70,71 The initial goal was to achieve a set of complementary models. For that, two unique molecules were selected as templates for the calculation of each of the six pharmacophore models (M1 – M6), respectively, developed in this study (detailed information is given in the Supporting Information). Catalyst Model: The 18 training set compounds were the basis for calculating a HypoGen pharmacophore model72 within Catalyst 4.7. All ten calculated models were used to screen the training and test set compounds (Charts 2 and 3). The model which found the highest number of active hERG blockers from both datasets was selected for prospective virtual screening studies. Theoretical Validation. Two approaches were taken for the assessment of theoretical performance. M1 – M6 models were screened against the highly active ChEMBL and the decoy databases. For VS, the “pharmacophore-fit” scoring function in LigandScout was employed, with a maximum number of omitted features set to zero. Two performance descriptors were used to evaluate the overall performance of the models. The enrichment factor (EF), which measures the yield of actives proportionally to the ratio of actives in the database, was calculated using the equation71 𝐸𝐹 = (𝑇𝑃⁄𝑛)⁄(𝐴⁄𝑁) where TP is the number of truly active compounds found and n is the number of hits from the database search, A is the amount of actives in the database and N is the total number of database molecules. The second approach was the construction of a receiver operating characteristic (ROC)


31


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 32 of 51

curve and the calculation of the area under the ROC curve (ROC-AUC). The ROC curve combines the sensitivity (Se) and specificity (Sp) into a graph, displaying the increase of false positives that results with increased true positives, according to the equations below71 𝑆𝑒 = 𝑇𝑃⁄(𝑇𝑃 + 𝐹𝑁) 𝑆𝑝 = 𝑇𝑁⁄(𝑇𝑁 + 𝐹𝑃) where TP is the number of retrieved true positives, TN is the amount of rejected truly negative compounds, FP is the retrieved false positive compounds, and FN is the number of false negative compounds. For a review on pharmacophore performance descriptors see Seidel et al., 2010.71 For the Catalyst model evaluation, the initial database was divided into training and test sets. Both sets were virtually screened in Discovery Studio using the best flexible search algorithm, with fit calculations computed using the “best fit” mode. As a final step, a retrospective validation of all models was performed using a previously published external dataset of know hERG blockers.51 Virtual screening was carried out as described above. Prospective Virtual Screening and Selection of Test Compounds. SPECS SP, SPECS NP and In-house databases were screened using the same settings as for the theoretical validation. After clustering and visual inspection of VH covering all pharmacophore models, compounds were selected for biological testing based on chemical diversity. Compounds were either obtained from in-house libraries or purchased from SPECS. Stock solutions (3-10 mM) were prepared in dimethylsulfoxide (DMSO) and stored at -20°C until further use. The identity and purity of compounds were determined by means of thin-layer chromatography (TLC), high performance liquid chromatography-mass spectrometry (HPLC-MS) and differential scanning calorimetry (DSC) analyses. All biologically tested compounds showed purity ≥ 95%.


32

Page 33 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Experimental Validation. Initial hERG Block Screening in Oocytes. Preparation of stage V–VI oocytes from Xenopus laevis (NASCO, Fort Atkinson, WI, USA), synthesis of capped runoff complementary ribonucleic

acid

(cRNA)

transcripts

from

linearized

complementary

deoxyribonucleic acid (cDNA) templates, and injection of cRNA were performed as described previously.73 hERG cDNAs were kindly provided by Dr. Sanguinetti (University of Utah, UT, USA). Currents through hERG channels were studied 1-3 days after cRNA injection using a twomicroelectrode voltage clamp technique with a Turbo TEC-03X amplifier (npi electronic GmbH, Tamm, Germany). The extracellular recording solution contained: 96 mM NaCl, 2 mM KCl, 1 mM MgCl2, 5mM HEPES and 1.8 mM CaCl2 (pH 7.5). Voltage-recording and current-injecting microelectrodes were filled with 3 M KCl and had resistances between 0.5 and 2 MΩ. Endogenous currents did not exceed 0.2 µA. Currents >3 µA were discarded to minimize voltage clamp errors. A precondition for all measurements was the achievement of stable peak current amplitudes over periods of 10 min after an initial run-up period. Stocks were diluted in extracellular solution on the day of each experiment, and the maximal DMSO concentration (1%) did not affect hERG currents. All compounds (30 µM) were applied by means of a fast perfusion system (ScreeningTool, npi electronic GmbH, Tamm, Germany).52 Thioridazine hydrochloride (SigmaAldrich GmbH, Vienna, Austria) was used as positive control. The pClamp software package version 10.1 (Molecular Devices, Sunnyvale, CA, USA) was used for data acquisition. Cell Culture. Human embryonic kidney (HEK 293) cells stably expressing hERG (a kind gift from Dr. Craig January, University of Wisconsin-Madison, WI, USA) were cultured in minimum essential media (MEM) (Life Technologies, Vienna, Austria), containing 10% fetal bovine serum (Life Technologies, Vienna, Austria), 400 µg/ml G418 (Eubio, Vienna, Austria) and 100 U/ml penicillin–streptomycin (Sigma-Aldrich GmbH, Vienna, Austria), at 37°C in an atmosphere of 5%


33


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 34 of 51

CO2, and 95% air. Before electrophysiological measurements, cells were harvested from their culture flasks using TrypLE™ Express (Life Technologies, Vienna, Austria) and centrifuged at 1000 rpm for 4 min. The pellet was then resuspended in the extracellular solution and directly used for electrophysiological recording. Whole-Cell Planar Patch Clamp. Currents through hERG channels stably expressed in HEK 293 cells were studied within 8h of harvest in the whole-cell configuration of the planar patch clamp technique (NPC-16 Patchliner®, Nanion Technologies GmbH, Munich, Germany), using EPC 10 patch clamp amplifier (HEKA Elektronik Dr. Schulze GmbH, Lambrecht/Pfalz, Germany).53,54 Currents were low-pass filtered at 10 kHz using the internal Bessel filter and sampled at 25 kHz. The extracellular solution contained: 140 mM NaCl, 4 mM KCl, 2 mM CaCl2, 1 mM MgCl2, 5 mM D-Glucose and 10 mM HEPES (pH 7.4) (Sigma-Aldrich GmbH, Vienna, Austria). The intracellular solution contained: 50 mM KCl, 10 mM NaCl, 60 mM KF, 20 mM EGTA and 10 mM HEPES (pH 7.2) (Sigma-Aldrich GmbH, Vienna, Austria). Compounds solutions were applied by means of the automated planar patch clamp platform NPC-16 Patchliner®. The PatchMaster software version 2.65 (HEKA Elektronik Dr. Schulze GmbH, Lambrecht/Pfalz, Germany) was used for data acquisition. Voltage Protocol. For the electrophysiological studies with both systems, stable peak current amplitudes over 10 minutes after an initial run-up phase was a precondition for the measurements. The voltage protocol was designed to simulate voltage changes during a cardiac action potential with a 300 ms depolarization to + 20 mV (analogous to plateau phase), a repolarization for 300 ms to − 40 mV (inducing a tail current) (− 50 mV for whole-cell patch clamp) and a final step to the holding potential (− 100 mV). The + 20 mV depolarization rapidly inactivates hERG channels, thereby limiting the amount of outward current. During the repolarization to – 40 / − 50 mV, the


34

Page 35 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


previously activated channels open due to rapid recovery from inactivation. The decreases in the resulting tail current amplitudes were taken as a measure of block development during a pulse train. Data Analysis. Origin software version 7.0 (OriginLab Corp., Northampton, MA, USA) was employed for data analysis and curve fitting. The cumulative concentration-inhibition curves were fitted using the Hill equation: 𝐼ℎ𝐸𝑅𝐺,𝑑𝑟𝑢𝑔 𝐼ℎ𝐸𝑅𝐺,𝑐𝑜𝑛𝑡𝑟𝑜𝑙

=

100 − 𝐴 𝐶 𝑛𝐻 1 + (𝐼𝐶 ) 50

+𝐴

in which IC50 is the concentration at which hERG inhibition is half-maximal, C is the applied drug concentration, A is the fraction of hERG current that is not blocked, and nH is the Hill coefficient.74 Data are presented as mean ± standard error (SE) of at least three oocytes from two different batches, or three independent measurements with HEK 293 cells.

ASSOCIATED CONTENT Supporting information. A chart of all 50 compounds selected for biological testing along with fit scores and detailed hERG screening data, additional information on M1 – M6 models development, Catalyst model training set screening results, 3D scatter plots depicting the drug chemical property space of novel hERG blockers, analysis full experimental procedures for synthesis of compounds 62 – S3 – S4 – S5 – S12 – S24 – S36 – S37. This material is available free of charge via the Internet at http://pubs.acs.org.


35


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 36 of 51

AUTHOR INFORMATION * Corresponding Authors Phone: 0043 512 507 58407 Fax: 0043 512 507 58499. E-mail: [email protected]. Phone: 0043 512 507 58253 Fax: 0043 512 507 58299. E-mail: [email protected]. Author Contributions The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript. Funding Sources This research was supported by a Marie Curie International Research Staff Exchange Scheme Fellowship within the 7th European Community Framework Programme (hERGscreen, 295174). D.S. thanks the University of Innsbruck for her position in the Erika Cremer Habilitation Program and for a young talents grant. This study was supported by FWF grant P22395 to S.H. Financial support by the graduate school program MolTag (Austrian Science Fund FWF, grant W1231) to P.S. is gratefully acknowledged. Notes The authors declare no competing financial interest. ACKNOWLEDGMENT We thank Prof. Ronald Gust, Dr. Gerhard Pürstinger, and Prof. Ulrich Griesser for providing the in-house compounds for biological testing, Stefan M. Noha for technical assistance, Philipp Schuster for help with the manuscript preparation, and J. Theiner for carrying out elemental analyses.


36

Page 37 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


ABBREVIATIONS AR, aromatic ring feature; AUC, area under the receiver operating characteristic curve; DSC, differential scanning calorimetry; EF, enrichment factor; EXVOLs, exclusion volume spheres; HBA, hydrogen bond acceptor; HBD, hydrogen bond donor; HEK 293, human embryonic kidney cells; HPFs, hydrophobic features; MEM, minimum essential media; PIF, positive ionizable feature; ROC, receiver operating characteristic; SPECS NP, SPECS natural products database; SPECS SP, SPECS synthetic compounds database; TdP, torsades de pointes; VH, virtual hits; VS, virtual screening.

REFERENCES 1. Yap, Y. G.; Camm, A. J. Drug induced QT prolongation and torsades de pointes. Heart 2003, 89, 1363-1372. 2. Sanguinetti, M. C.; Tristani-Firouzi, M. hERG potassium channels and cardiac arrhythmia. Nature 2006, 440, 463-469. 3. Vandenberg, J. I.; Perry, M. D.; Perrin, M. J.; Mann, S. A.; Ke, Y.; Hill, A. P. hERG K(+) channels: structure, function, and clinical significance. Physiol. Rev. 2012, 92, 1393-1478. 4. Redfern, W. S.; Carlsson, L.; Davis, A. S.; Lynch, W. G.; MacKenzie, I.; Palethorpe, S.; Siegl, P. K.; Strang, I.; Sullivan, A. T.; Wallis, R.; Camm, A. J.; Hammond, T. G. Relationships between preclinical cardiac electrophysiology, clinical QT interval prolongation and torsade de pointes for a broad range of drugs: evidence for a provisional safety margin in drug development. Cardiovasc. Res. 2003, 58, 32-45. 5. International Conference on Harmonisation. Guidance on S7B nonclinical evaluation of the potential for delayed ventricular repolarization (QT Interval Prolongation) by human pharmaceuticals. 2005, 1-9. 6. Raschi, E.; Ceccarini, L.; De Ponti, F.; Recanatini, M. hERG-related drug toxicity and models for predicting hERG liability and QT prolongation. Expert Opin. Drug Metab. Toxicol. 2009, 5, 1005-1021. 7. Chi, K. R. Revolution dawning in cardiotoxicity testing. Nat. Rev. Drug Discov. 2013, 12, 565-567. 8. Dennis, A.; Wang, L.; Wan, X.; Ficker, E. hERG channel trafficking: novel targets in drug-induced long QT syndrome. Biochem. Soc. Trans. 2007, 35, 1060-1063. 9. Durdagi, S.; Deshpande, S.; Duff, H. J.; Noskov, S. Y. Modeling of open, closed, and open-inactivated states of the hERG1 channel: structural mechanisms of the state-dependent drug binding. J. Chem. Inf. Model. 2012, 52, 2760-2774.


37


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 38 of 51

10. Vilums, M.; Overman, J.; Klaasse, E.; Scheel, O.; Brussee, J.; AP, I. J. Understanding of molecular substructures that contribute to hERG K+ channel blockade: synthesis and biological evaluation of E-4031 analogues. ChemMedChem 2012, 7, 107-113. 11. Lawrence, C. L.; Pollard, C. E.; Hammond, T. G.; Valentin, J. P. In vitro models of proarrhythmia. Br. J. Pharmacol. 2008, 154, 1516-1522. 12. Tan, Y.; Chen, Y.; You, Q.; Sun, H.; Li, M. Predicting the potency of hERG K(+) channel inhibition by combining 3D-QSAR pharmacophore and 2D-QSAR models. J. Mol. Model. 2012, 18, 1023-1036. 13. Su, B. H.; Shen, M. Y.; Esposito, E. X.; Hopfinger, A. J.; Tseng, Y. J. In silico binary classification QSAR models based on 4D-fingerprints and MOE descriptors for prediction of hERG blockage. J. Chem. Inf. Model. 2010, 50, 1304-1318. 14. Polak, S.; Wiśniowska, B.; Ahamadi, M.; Mendyk, A. Prediction of the hERG potassium channel inhibition potential with use of artificial neural networks. Applied Soft. Computing 2011, 11, 2611-2617. 15. Ekins, S.; Crumb, W. J.; Sarazan, R. D.; Wikel, J. H.; Wrighton, S. A. Three-dimensional quantitative structure-activity relationship for inhibition of human ether-a-go-go-related gene potassium channel. J. Pharmacol. Exp. Ther. 2002, 301, 427-434. 16. Rayan, A.; Falah, M.; Raiyn, J.; Da'adoosh, B.; Kadan, S.; Zaid, H.; Goldblum, A. Indexing molecules for their hERG liability. Eur. J. Med. Chem. 2013, 65, 304-314. 17. Cavalli, A.; Poluzzi, E.; De Ponti, F.; Recanatini, M. Toward a Pharmacophore for Drugs Inducing the Long QT Syndrome: Insights from a CoMFA Study of HERG K+ Channel Blockers. J. Med. Chem. 2002, 45, 3844-3853. 18. Durdagi, S.; Duff, H. J.; Noskov, S. Y. Combined receptor and ligand-based approach to the universal pharmacophore model development for studies of drug blockade to the hERG1 pore domain. J. Chem. Inf. Model. 2011, 51, 463-474. 19. Johnson, S. R.; Yue, H.; Conder, M. L.; Shi, H.; Doweyko, A. M.; Lloyd, J.; Levesque, P. Estimation of hERG inhibition of drug candidates using multivariate property and pharmacophore SAR. Bioorg. Med. Chem. 2007, 15, 6182-6192. 20. Leong, M. K. A novel approach using pharmacophore ensemble/support vector machine (PhE/SVM) for prediction of hERG liability. Chem. Res. Toxicol. 2007, 20, 217-226. 21. Yamakawa, Y.; Furutani, K.; Inanobe, A.; Ohno, Y.; Kurachi, Y. Pharmacophore modeling for hERG channel facilitation. Biochem. Biophys. Res. Commun. 2012, 418, 161-166. 22. Coi, A.; Bianucci, A. M. Combining structure- and ligand-based approaches for studies of interactions between different conformations of the hERG K(+) channel pore and known ligands. J. Mol. Graph. Model. 2013, 46, 93-104. 23. Wang, M.; Yang, X.-G.; Xue, Y. Identifying hERG potassium channel inhibitors by machine learning methods. QSAR Comb. Sci. 2008, 27, 1028-1035. 24. Wang, S.; Li, Y.; Wang, J.; Chen, L.; Zhang, L.; Yu, H.; Hou, T. ADMET evaluation in drug discovery. 12. Development of binary classification models for prediction of hERG potassium channel blockage. Mol. Pharm. 2012, 9, 996-1010. 25. Zemzemi, N.; Bernabeu, M. O.; Saiz, J.; Cooper, J.; Pathmanathan, P.; Mirams, G. R.; Pitt-Francis, J.; Rodriguez, B. Computational assessment of drug-induced effects on the electrocardiogram: from ion channel to body surface potentials. Br. J. Pharmacol. 2013, 168, 718-733. 26. Wang, S.; Li, Y.; Xu, L.; Li, D.; Hou, T. Recent developments in computational prediction of hERG blockage. Curr. Top. Med. Chem. 2013, 13, 1317-1326.


38

Page 39 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


27. Broccatelli, F.; Mannhold, R.; Moriconi, A.; Giuli, S.; Carosati, E. QSAR modeling and data mining link Torsades de Pointes risk to the interplay of extent of metabolism, active transport, and HERG liability. Mol. Pharm. 2012, 9, 2290-2301. 28. Doddareddy, M. R.; Klaasse, E. C.; Shagufta; Ijzerman, A. P.; Bender, A. Prospective validation of a comprehensive in silico hERG model and its applications to commercial compound and drug databases. ChemMedChem 2010, 5, 716-729. 29. Beattie, K. A.; Luscombe, C.; Williams, G.; Munoz-Muriedas, J.; Gavaghan, D. J.; Cui, Y.; Mirams, G. R. Evaluation of an in silico cardiac safety assay: using ion channel screening data to predict QT interval changes in the rabbit ventricular wedge. J. Pharmacol. Toxicol. Methods 2013, 68, 88-96. 30. Czodrowski, P. hERG me out. J. Chem. Inf. Model. 2013, 53, 2240-2251. 31. Gavaghan, C. L.; Arnby, C. H.; Blomberg, N.; Strandlund, G.; Boyer, S. Development, interpretation and temporal evaluation of a global QSAR of hERG electrophysiology screening data. J. Comput. Aided Mol. Des. 2007, 21, 189-206. 32. Ekins, S.; Mestres, J.; Testa, B. In silico pharmacology for drug discovery: applications to targets and beyond. Br. J. Pharmacol. 2007, 152, 21-37. 33. Schuster, D. 3D pharmacophores as tools for activity profiling. Drug. Discov. Today Technol. 2010, 7, e203-270. 34. Sanders, M. P.; Barbosa, A. J.; Zarzycka, B.; Nicolaes, G. A.; Klomp, J. P.; de Vlieg, J.; Del Rio, A. Comparative analysis of pharmacophore screening tools. J. Chem. Inf. Model. 2012, 52, 1607-1620. 35. Spitzer, G. M.; Heiss, M.; Mangold, M.; Markt, P.; Kirchmair, J.; Wolber, G.; Liedl, K. R. One concept, three implementations of 3D pharmacophore-based virtual screening: distinct coverage of chemical search space. J. Chem. Inf. Model. 2010, 50, 1241-1247. 36. Du-Cuny, L.; Chen, L.; Zhang, S. A critical assessment of combined ligand- and structure-based approaches to HERG channel blocker modeling. J. Chem. Inf. Model. 2011, 51, 2948-2960. 37. Catalyst software package, Vers. 4.7, Accelrys Software Inc.: San Diego, CA, USA, 2003. 38. Schuster, D.; Waltenberger, B.; Kirchmair, J.; Distinto, S.; Markt, P.; Stuppner, H.; Rollinger, J. M.; Wolber, G. Predicting cyclooxygenase inhibition by three-dimensional pharmacophoric profiling. Part I: Model generation, validation and applicability in ethnopharmacology. Mol. Inf. 2010, 29, 75-86. 39. Wolber, G.; Langer, T. LigandScout: 3-D pharmacophores derived from protein-bound ligands and their use as virtual screening filters. J. Chem. Inf. Model. 2005, 45, 160-169. 40. Crumb, W. J., Jr. Loratadine blockade of K+ channels in human heart: comparison with terfenadine under physiological conditions. J. Pharmacol. Exp. Ther. 2000, 292, 261-264. 41. Katayama, Y.; Fujita, A.; Ohe, T.; Findlay, I.; Ruachi, Y. Inhibitory effects of vesnarinone on cloned cardiac delayed rectifier K+ channels expressed in a mammalian cell line. J. Pharmacol. Exp. Ther. 2000, 294, 339-346. 42. Kim, K.-S.; Kim, E.-J. The phenothiazine drugs inhibit hERG potassium channels. Drug Chem. Toxicol. 2005, 28, 303-313. 43. Kuryshev, Y. A.; Brown, A. M.; Wang, L.; Benedict, C. R.; Rampe, D. Interactions of the 5-hydroxytryptamine 3 antagonist class of antiemetic drugs with human cardiac ion channels. J. Pharmacol. Exp. Ther. 2000, 295, 614-620.


39


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 40 of 51

44. Sanchez-Chapula, J. A.; Ferrer, T.; Navarro-Polanco, R. A.; Sanguinetti, M. C. Voltagedependent profile of human Ether-a-go-go-related gene channel block is influenced by a single residue in the S6 transmembrane domain. Mol. Pharmacol. 2003, 63, 1051-1058. 45. Teschemacher, A. G.; Seward, E. P.; Hancox, J. C.; Witchel, H. J. Inhibition of the current of heterologously expressed HERG potassium channels by imipramine and amitriptyline. Br. J. Pharmacol. 1999, 128, 479-485. 46. Tie, H.; Walker, B. D.; Singleton, C. B.; Valenzuela, S. M.; Bursill, J. A.; Wyse, K. R.; Breit, S. N.; Campbell, T. J. Inhibition of HERG potassium channels by the antimalarial agent halofantrine. Br. J. Pharmacol. 2000, 130, 1967-1975. 47. Walker, B. D.; Valenzuela, S. M.; Singleton, C. B.; Tie, H.; Bursill, J. A.; K.R., W.; Qiu, M. R.; Breit, S. N.; Campbell, T. J. Inhibition of hERG channels stably expressed in a mammalian cell line by the antianginal agent perhexiline maleate. Br. J. Pharmacol. 1999, 127, 243-251. 48. Zhang, S.; Rajamani, S.; Chen, Y.; Gong, Q.; Rong, Y.; Zhou, Z.; Ruoho, A.; January, C. T. Cocaine blocks HERG, but not KvLQT1+minK, potassium channels. Mol. Pharmacol. 2001, 59, 1069-1076. 49. Zhang, S.; Zhou, Z.; Gong, Q.; Makielski, J. C.; January, C. T. Mechanism of block and identification of the verapamil binding domain to HERG potassium channels. Circ. Res. 1999, 84, 989-998. 50. Aronov, A. M. Predictive in silico modeling for hERG channel blockers. Drug Discov. Today 2005, 10, 149-155. 51. de Bruin, M. L.; Pettersson, M.; Meyboon, R. H. B.; Hoes, A. W.; Leufkens, H. G. M. Anti-HERG activity and the risk of drug-induced arrhythmias and sudden death. Eur. Heart J. 2005, 26, 590-597. 52. Baburin, I.; Beyl, S.; Hering, S. Automated fast perfusion of Xenopus oocytes for drug screening. Pflugers Arch. 2006, 453, 117-123. 53. Milligan, C. J.; Li, J.; Sukumar, P.; Majeed, Y.; Dallas, M. L.; English, A.; Emery, P.; Porter, K. E.; Smith, A. M.; McFadzean, I.; Beccano-Kelly, D.; Bahnasi, Y.; Cheong, A.; Naylor, J.; Zeng, F.; Liu, X.; Gamper, N.; Jiang, L. H.; Pearson, H. A.; Peers, C.; Robertson, B.; Beech, D. J. Robotic multiwell planar patch-clamp for native and primary mammalian cells. Nat. Protoc. 2009, 4, 244-255. 54. Polonchuk, L. Toward a new gold standard for early safety: Automated temperaturecontrolled hERG test on the PatchLiner. Front. Pharmacol. 2012, 3, 1-7. 55. Jenks, C. W. Extraction studies of Tabernanthe iboga and Voacanga africana. Nat. Prod. Lett. 2002, 16, 71-76. 56. Koenig, X.; Kovar, M.; Rubi, L.; Mike, A. K.; Lukacs, P.; Gawali, V. S.; Todt, H.; Hilber, K.; Sandtner, W. Anti-addiction drug ibogaine inhibits voltage-gated ionic currents: A study to assess the drug's cardiac ion channel profile. Toxicol. Appl. Pharmacol. 2013, 273, 259268. 57. Vuorinen, A.; Nashev, L. G.; Odermatt, A.; Rollinger, J. M.; Schuster, D. Pharmacophore model refinement for 11-hydroxysteroid dehydrogenase inhibitors: Search for modulators of intracellular glucocorticoid concentrations. Mol. Inf. 2014, 33, 15-25. 58. Klon, A. E. Machine learning algorithms for the prediction of hERG and CYP450 binding in drug development. Exp. Opin. Drug Metab. Toxicol. 2010, 6, 821-833. 59. Hawkins, P. C.; Nicholls, A. Conformer generation with OMEGA: learning from the data set and the analysis of failures. J. Chem. Inf. Model. 2012, 52, 2919-2936.


40

Page 41 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


60. Hawkins, P. C.; Skillman, A. G.; Warren, G. L.; Ellingson, B. A.; Stahl, M. T. Conformer generation with OMEGA: algorithm and validation using high quality structures from the Protein Databank and Cambridge Structural Database. J. Chem. Inf. Model. 2010, 50, 572-584. 61. OMEGA, version 2.3.3; OpenEye Scientific Software, I.: Santa Fe, NM, USA, 20092013. 62. Gaulton, A.; Bellis, L. J.; Bento, A. P.; Chambers, J.; Davies, M.; Hersey, A.; Light, Y.; McGlinchey, S.; Michalovich, D.; Al-Lazikani, B.; Overington, J. P. ChEMBL: a large-scale bioactivity database for drug discovery. Nucleic Acids Res. 2012, 40, D1100-D1107. 63. Dunlop, J.; Bowlby, M.; Peri, R.; Vasilyev, D.; Arias, R. High-throughput electrophysiology: an emerging paradigm for ion-channel screening and physiology. Nat. Rev. Drug Discov. 2008, 7, 358-368. 64. Coon, T.; Moree, W. J.; Li, B.; Yu, J.; Zamani-Kord, S.; Malany, S.; Santos, M. A.; Hernandez, L. M.; Petroski, R. E.; Sun, A.; Wen, J.; Sullivan, S.; Haelewyn, J.; Hedrick, M.; Hoare, S. J.; Bradbury, M. J.; Crowe, P. D.; Beaton, G. Brain-penetrating 2-aminobenzimidazole H(1)-antihistamines for the treatment of insomnia. Bioorg. Med. Chem. Lett. 2009, 19, 43804384. 65. Lin, H.; Yamashita, D. S.; Xie, R.; Zeng, J.; Wang, W.; Leber, J.; Safonov, I. G.; Verma, S.; Li, M.; Lafrance, L.; Venslavsky, J.; Takata, D.; Luengo, J. I.; Kahana, J. A.; Zhang, S.; Robell, K. A.; Levy, D.; Kumar, R.; Choudhry, A. E.; Schaber, M.; Lai, Z.; Brown, B. S.; Donovan, B. T.; Minthorn, E. A.; Brown, K. K.; Heerding, D. A. Tetrasubstituted pyridines as potent and selective AKT inhibitors: Reduced CYP450 and hERG inhibition of aminopyridines. Bioorg. Med. Chem. Lett. 2010, 20, 684-688. 66. Moree, W. J.; Jovic, F.; Coon, T.; Yu, J.; Li, B. F.; Tucci, F. C.; Marinkovic, D.; Gross, R. S.; Malany, S.; Bradbury, M. J.; Hernandez, L. M.; O'Brien, Z.; Wen, J.; Wang, H.; Hoare, S. R.; Petroski, R. E.; Sacaan, A.; Madan, A.; Crowe, P. D.; Beaton, G. Novel benzothiophene H1antihistamines for the treatment of insomnia. Bioorg. Med. Chem. Lett. 2010, 20, 2316-2320. 67. Webb, R. L.; Schiering, N.; Sedrani, R.; Maibaum, J. Direct renin inhibitors as a new therapy for hypertension. J. Med. Chem. 2010, 53, 7490-7520. 68. Park, S. J.; Buschmann, H.; Bolm, C. Bioactive sulfoximines: syntheses and properties of Vioxx analogs. Bioorg. Med. Chem. Lett. 2011, 21, 4888-4890. 69. Schuster, D.; Laggner, C.; Steindl, T. M.; Palusczak, A.; Hartmann, R. W.; Langer, T. Pharmacophore modeling and in silico screening for new P450 19 (aromatase) inhibitors. J. Chem. Inf. Model. 2006, 46, 1301-1311. 70. Wolber, G.; Dornhofer, A. A.; Langer, T. Efficient overlay of small organic molecules using 3D pharmacophores. J. Comput. Aided Mol. Des. 2006, 20, 773-788. 71. Seidel, T.; Ibis, G.; Bendix, F.; Wolber, G. Strategies for 3D pharmacophore-based virtual screening. Drug Discov. Today Techn. 2010, 7, e221-e228. 72. Li, H.; Sutter, J.; Hoffmann, R. HypoGen: An Automated System for Generating 3D Predictive Pharmacophore Models. In Pharmacophore perception, development, and use in drug design, Güner, O. F., Ed. International University Line, La Jolla, CA: 2000; pp 172-189. 73. Stork, D.; Timin, E. N.; Berjukow, S.; Huber, C.; Hohaus, A.; Auer, M.; Hering, S. State dependent dissociation of HERG channel inhibitors. Br. J. Pharmacol. 2007, 151, 1368-1376. 74. Windisch, A.; Timin, E.; Schwarz, T.; Stork-Riedler, D.; Erker, T.; Ecker, G.; Hering, S. Trapping and dissociation of propafenone derivatives in HERG channels. Br. J. Pharmacol. 2011, 162, 1542-1552.


41


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 42 of 51

FOR TABLE OF CONTENTS USE ONLY

Experimentally validated hERG pharmacophore models as valuable cardiotoxicity prediction tools Jadel M. Kratz, Daniela Schuster, Michael Edtbauer, Priyanka Saxena, Christina E. Mair, Julia Kirchebner, Barbara Matuszczak, Igor Baburin, Steffen Hering, Judith M. Rollinger


42

Page 43 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Representative structures of drugs withdrawn from the market due to hERG-related QT interval prolongation and severe risk of fatal arrhythmias



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Training compounds used for Catalyst model generation


Page 44 of 51

Page 45 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Test compounds used for Catalyst model theoretical validation



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

General workflow employed in this study


Page 46 of 51

Page 47 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Catalyst pharmacophore model for hERG blockers. The distances between the features are indicated in Å. Chemical features are color-coded: positively ionizable – red, hydrophobic – cyan, aromatic hydrophobic – blue. 190x142mm (300 x 300 DPI)



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Training compounds and optimized ligand-based LigandScout pharmacophore models (M1 – M6). Chemical features of the models are color-coded: hydrophobic – yellow; hydrogen bond acceptor – red; aromatic ring – blue parallel rings; positively ionizable group – blue star; exclusion volumes – grey.


Page 48 of 51

Page 49 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


ROC plots of screening the highly active database and decoy set using the models M1 – M6 in parallel (ROCAUC = 0.91) (A) and using the Catalyst model (ROC-AUC = 0.89) (B). 181x250mm (269 x 269 DPI)



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Inhibition by compounds (30 µM) of hERG current in the oocyte two micro electrode assay, given as mean ± SE (n = 3-5). Thioridazine (8) was used as positive control. 176x98mm (300 x 300 DPI)


Page 50 of 51

Page 51 of 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Concentration-response curves of hERG current inhibition by compounds in the HEK293 cells patch clamp assay (n = 3-5). Only compounds that showed ≥ 50% block in the preliminary screening were selected for this further validation step. 254x339mm (229 x 229 DPI)


Experimentally Validated hERG Pharmacophore Models as

Recommend Documents